Vague prompts and “vibe coding” can leave teams with code that’s hard to trust, review, or evolve—especially once a codebase is already in motion. This tutorial walks through OpenSpec, a spec-driven toolkit that aims to align humans and AI assistants on what to build before implementation, with a workflow designed for ongoing development and changes.

Key takeaways

• How OpenSpec structures work into proposals, tasks, and spec updates to keep scope changes auditable and transparent
• How it separates “current” specs from proposed updates to handle evolving systems more cleanly
• A hands-on setup: installing via npm, initializing in a project, and generating project context files for an AI agent
• Moving from a change proposal to implementation while keeping results more deterministic and reviewable